Although Elon Musk still believes that in 2020 Telsa will have a car on the road in which the driver can safely fall asleep at the wheel, almost no one else in the autonomous vehicle (AV) world is so sanguine. Readers of my blog know that for sometime I thought that there was too much optimism about the imminent advent of AVs; however, even I was surprised by the discouraged tone of a recent Design News article on the status of AVs. The most shocking quote to me was from John Krafcik, CEO of Google’s self-driving car unit, Waymo, as stated in the Design News article:
"It’s really, really hard," Krafcik said during a live-streamed tech conference. "You don’t know what you don’t know until you’re actually in there and trying to do things." Krafcik went on to say that the auto industry might never produce a car capable of driving at any time of year, in any weather, under any conditions. "Autonomy will always have some constraints," he added.
Wow! A few years ago, many were lamenting near term loss of jobs for taxi, Uber, truck drivers and the like. What has surprised me about all of this past over optimism is that many experts in the AV technology were cautious years ago. Steven Shladover’s June 2016 article in Scientific American suggested that full AVs would not be available until well beyond 2040, perhaps not until 2075. It is surprising to me that leaders in the auto industry did not take the caution of experts like Shladover more seriously. A quote from Shladover’s article is most telling:
“The auto industry and press have oversold the automated car. Simple road encounters pose huge challenges for computers, and robotic chauffeurs are decades away.”
Three examples of simple challenges for humans that are very hard for AV’s might be:
- A delivery truck is partly blocking the road after suffering a mechanical breakdown. The human truck driver has called for help. An AV approaches from behind the truck and stops. There is no traffic coming either way so the truck driver waves the AV to go around him: however, to pass the truck the AV must drive over a double yellow line in the highway. Since this operation is against the law, the AV will not pass and waits there for two hours until the truck is towed away.
- The wind blows a piece of cardboard on to the road. A human quickly recognizes the cardboard as something that can be driven over. An AV doesn’t necessarily have this context and may stop if it can’t go around it.
- It is 30º and a light rain is falling. Most people recognize that water on a bridge will freeze before it does on the road surface. An AV might not recognize this fact and skid on a bridge. Weather in general is a great challenge for AVs.
Another trial for AVs is that working through the intermediate stages of autonomous driving might be dangerous to the driver. SAE Level 3 autonomy is such that the AV does much of the driving with human backup. The problem here is that the human can become bored or disengaged and not respond when needed. My wife just bought a 2019 auto. In 2016, the auto company was suggesting they would have partial AVs in the near future. Today, this feature was not an option on any car we looked at. I believe this concern is the reason. It may be that only full automation is safe.
Why do AVs and artificial intelligence (AI) have so much difficulty in performing simple human tasks? I think one reason is lack of context. Humans are able to do what we consider simple tasks because we have lived in the world and have developed a sophisticated, yet taken for granted, context of how the world works. AVs and AIs struggle because this context of how the world works is not built into their software. The three examples I mentioned above are simple for humans to solve because of years of developed context. The bottom line is that the skills of a human are profoundly difficult to build into machines. I believe the future will point out this fact over and over again.
The good news for those of us that assemble electronics is that more sensors and electronics will be needed than first thought as we march more slowly toward AVs. Even today, an auto is becoming a computer, or multiple computers on wheels. So there are sanguine days ahead for this of us in automotive electronics.